[Bug c/106927] New: false-positive -Werror=restrict warnings for memmove (calling memcpy) within array

[Bug c/106927] New: false-positive -Werror=restrict warnings for memmove (calling memcpy) within array

[gcc at boris dot fau.re]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106927

            Bug ID: 106927
           Summary: false-positive -Werror=restrict warnings for memmove
                    (calling memcpy) within array
           Product: gcc
           Version: 12.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gcc at boris dot fau.re
  Target Milestone: ---

When compiling the following code with -Werror=restrict, gcc complains:
$cat foo.c
#include <string.h>

int main (void)
{
    char *s = "abcdefghijklmnopqrstuvwxyz012345";
    size_t d = strlen(s);

    __builtin_memmove(s, s + 1, d - 1);

    return 0;
}
$ gcc -c -O2 -Werror=restrict -o foo.o foo.c
foo.c: In function 'main':
foo.c:8:5: error: '__builtin_memcpy' accessing 31 bytes at offsets 0 and 1
overlaps 30 bytes at offset 1 [-Werror=restrict]
    8 |     __builtin_memmove(s, s + 1, d - 1);
      |     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

This was reproduced on gentoo with gcc-11.3.0, on RHEL9 with gcc-11.2.1, on
debian bullseye with gcc-12.2.0, on debian buster with gcc-9.5.0, RHEL8 with
gcc-8.5.0.

Please note that, on gcc-12.2.0, without -O2, the compilation succeeds.

[Bug c/106927] false-positive -Werror=restrict warnings for memmove (calling memcpy) within array

[gcc at boris dot fau.re]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106927

--- Comment #1 from Boris Faure <gcc at boris dot fau.re> ---
The code is indeed incorrect.

It should be:
#include <string.h>

int main (void)
{
    char s[] = "abcdefghijklmnopqrstuvwxyz012345";
    size_t d = strlen(s);

    __builtin_memmove(s, s + 1, d - 1);

    return 0;
}

However, the error message is not helping.

[Bug tree-optimization/106927] false-positive -Werror=restrict warnings for memmove (calling memcpy) within string literal

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106927

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
           Keywords|                            |diagnostic
     Ever confirmed|0                           |1
                 CC|                            |msebor at gcc dot gnu.org
   Last reconfirmed|                            |2022-09-13
          Component|c                           |tree-optimization
            Summary|false-positive              |false-positive
                   |-Werror=restrict warnings   |-Werror=restrict warnings
                   |for memmove (calling        |for memmove (calling
                   |memcpy) within array        |memcpy) within string
                   |                            |literal

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
We issue the diagnostic for

__builtin_memcpy ("abcdefghijklmnopqrstuvwxyz012345", &MEM <char[33]> [(void
*)"abcdefghijklmnopqrstuvwxyz012345" + 1B], 31);

but we'd better issue a -Wwrite-strings diagnostic.

The actually emitted diagnostic is indeed somewhat misleading but the
reason is we simplify the memmove call to memcpy as seen above because
we have

      if (code == BUILT_IN_MEMMOVE)
        {
          /* Both DEST and SRC must be pointer types.
             ??? This is what old code did.  Is the testing for pointer types
             really mandatory?

             If either SRC is readonly or length is 1, we can use memcpy.  */
          if (!dest_align || !src_align)
            return false;
          if (readonly_data_expr (src)
              || (tree_fits_uhwi_p (len)
                  && (MIN (src_align, dest_align) / BITS_PER_UNIT
                      >= tree_to_uhwi (len))))
            {
              tree fn = builtin_decl_implicit (BUILT_IN_MEMCPY);
...

that should probably suppress -Wrestrict on the call?